This paper proposes a conceptual model for LLM-mediated workflows that treats workflow definitions, instances, and context snapshots as persistent knowledge objects within a shared substrate.

  • The model is Lisp-inspired but language-independent, using symbolic forms and object identity as explanatory lenses rather than implementation commitments.
  • It establishes a central semantic distinction between 'derive' (deterministic computation over available state) and 'infer' (mediated LLM judgment under declared context and executor-controlled capability policy).
  • Workflows are represented as inspectable, resumable, and reviewable knowledge objects that produce and leave traces of knowledge.

The authors present this as a preliminary conceptual account of semantic persistence, noting that formal transition semantics remain future work.